Several studies have characterized personal exposure to RF-EMF, which allows possible effects on health to be studied. All equipment has a detection limit, below which we obtain nondetects or censored data. This problem is a challenge for researchers as it makes the analysis of such data complex. We suggest reconsidering the statistical protocols of the nondetects analysis by comparing four different methods. Three of them substitute censored data using different approaches: regression on order of statistics (ROS) to simulate data below the detection limit (Method 1), substituting nondetect values by the detection limit divided by 2 (Method 2), a naïve calculation (Method 3) using the detection limit as a valid measurement. The fourth method consists of considering censored data to be missing values (Method 4). This article examines how these methods affect the quantification of personal exposure. We considered data from 14 frequency bands from FM to WiMax measured in Albacete (Spain) for 76 days every 10 s by a personal exposimeter (PEM) Satimo EME Spy 140.Methods 3 and 2 gave similar mean and median values to Method 1, but both underestimated the mean values when high nondetects records occurred, which conditioned the physical description of the real situation. The mean values calculated by Method 4 differed from those obtained by Method 1 but were similar when the percentage of nondetects was below 20%.Our comparison suggests that nondetects can be neglected when the percentage of censored data is low to provide a more realistic physical situation.
Keywords: Censored data; Detection limit; Exposimeter; Personal exposure; Radiofrequency electromagnetic fields.